Production in Entrepreneurial Firms: The Effects of FinancialConstraints on Labor and Capital

Mark J. GarmaiseUCLA Anderson

Correspondence to: Mark Garmaise, UCLA Anderson, 110 Westwood Plaza, Los Angeles, CA, 90095. E-mail:

mark.garmaise@anderson.ucla.edu. I have benefitted from the suggestions and comments of Tony Bernardo, Matthias

Kahl, Kjell Nyborg, Robert L. McDonald, an anonymous referee and seminar participants at the University of

Southern California, Tel Aviv University and Hebrew University. I thank the Harold Price Center for Entrepreneurial

Studies for support.

Production in Entrepreneurial Firms: The Effects of Financial

Constraints on Labor and Capital

Abstract

We model the contrasting capital-labor decisions of financially constrained and unconstrained

firms. We show that financially restricted firms use relatively more labor than physical capital

because informed employees provide more efficient financing than uninformed capital suppliers.

We demonstrate that constrained firms cannot easily attract new employees to replace existing

staff. Their greater employee retention aligns owner-worker incentives and encourages workers to

make firm-specific investments. Constrained firms, however, gradually suffer from their inability

to replace low-quality workers, such that their relative labor productivity decreases over time.

Empirical tests utilizing instrumental variables confirm several implications of the theory.

Introduction

Small entrepreneurial firms and larger companies have very different labor management policies.

One important distinction is that small firms are much more labor-intensive in their production

processes. For example, the large firms in the Compustat data set in 1998 had a median total assets-

employee ratio of $189,628 while the median ratio for the small firms in the 1998 National Survey

of Small Business Finance (NSSBF) was $18,520.1 A second disparity is that survey and anecdotal

evidence consistently show that entrepreneurial work environments are more cooperative.2 What

explains these different uses of labor in small and large firms? Is there a connection between small

firms’ heavy reliance on labor and their collaborative modes of production? In this paper, we argue

that it is the financially constrained status of many small firms that drives both their relatively

high employment levels and their enlistment of employee firm-specific investments.

We develop a model in which there are both constrained and unconstrained entrepreneurs.

Financially restricted firms must seek vendor financing from their suppliers. We presume that

the supplier of labor (i.e., a worker) learns about the firm in the course of production, while the

supplier of physical capital (e.g., a vendor or a bank leasing equipment to the entrepreneur) does

not receive information until production is complete. The worker’s ability to learn in the course of

performing his duties enables him to exploit the real option to quit the firm if the entrepreneur’s

quality (e.g., the innovativeness of his project or the depth of his management skill) is revealed to

be low. Workers are thus willing to join risky new businesses, even without an initial compensatory

payment, because they will remain only with successful ventures. Suppliers of capital do not learn

the entrepreneur’s quality as quickly as workers do, and hence can only rent out their assets to firms

with the resources to offer some fee in advance. Constrained firms’ inability to offer their suppliers

cash in advance thus leads them to make greater use of labor rather than capital in production.

We then demonstrate that even financially restricted firms that become successful find it hard

to attract new employees (and thereby replace existing staff), and hence current workers can expect

to receive a larger share of future firm profits in these firms. This has the implication that in a

constrained firm a worker has incentives that are better aligned with the owner, with the result

that the employee is more willing to make firm-specific investments. These investments benefit the

firm such that constrained firms will exhibit strong performance early in their lives. Unconstrained1The Compustat firms had a median of 600 employees, while the NSSBF firms had a median of 5 employees.2Examples include Wilkinson (1999) and “Nine to Five - No Comparison: Having worked for both large and small

companies, These executives have come to the same conclusion: Small is better,” Wall Street Journal, May 22, 1997,p. R.10.

1

firms, however, benefit as they age from their ability to replace low quality workers. Our model

thus predicts that the labor productivity of financially constrained firms relative to unconstrained

firms will decrease over time.

We test several of the implications of the model. Measuring the effects of financial constraints

on a firm’s production and employment strategies requires a reasonable proxy for having limited

access to credit. Using data from the 1998 National Survey of Small Business Finance, we analyze

the owner and firm characteristics that are associated with loan rejections and then consider the

capital-labor ratios of firms that are more likely to have loan applications denied. We do not simply

regress capital-labor ratios on whether a firm’s application was denied, because firms with little

capital to serve as collateral might be more likely to be rejected, thereby generating an endogenous

relationship. Instead we show that local bank concentration, the owner’s home equity, net worth,

and ethnic status, and the firm’s credit score are all good predictors of bank loan application

rejection and hence serve as good instruments for financial constraints. We then show, in a two-

stage least squares analysis, that the instruments for bank loan rejection are associated with low

capital-labor ratios, controlling for firm size, age and industry, as predicted by our theory.

We further test the implication of the model that the relative productivity of constrained firms

will decrease with firm age. This is confirmed in empirical results that show that financially

restricted firms experience slower productivity growth, controlling for the level and type of assets

and the number of employees in the firm. We also provide evidence, using data from the 1996-1997

National Organizations Survey, that is consistent with our theoretical prediction that workers in

constrained firms are more likely to make firm-specific investments.

Our theory and empirical work suggest that financial constraints can have a substantial impact

on the investment decisions and growth trajectories of entrepreneurial firms. Small firms (those

with fewer than 500 employees) are responsible for roughly 58 percent of U.S. private employment

(Headd (2000)) and 50 percent of private, nonfarm GDP (Joel Popkin and Company (2001)).

Given this importance, their strategies for setting employment levels and deploying capital have

crucial implications for the overall allocation and organization of human and physical capital in the

economy.

The paper is organized as follows. Section 1 provides a literature review. We introduce the

basic model in Section 2 and provide results on the optimal capital-labor choices of financially

constrained and unconstrained firms. In Section 3 we extend the model to analyze the effects of

financial constraints on the firm-specific investment decisions made by employees, and we examine

2

the relative productivities of constrained and unconstrained firms over time. Section 4 describes

empirical tests that examine the relationship between financial constraints and capital-labor ratios

and productivity growth. In Section 5 we provide empirical work linking financial constraints to

firm-specific investments by workers. We conclude the paper in Section 6. Formal proofs of the

results are given in the Appendix.

1 Literature Review

Our work relates to several different streams of the literature. Hanka (1998), Ofek (1993), Kang

and Shivdasani (1997) and Bronars and Deere (1991) link firms’ capital structures to their labor

negotiations strategies and firing policies. The thrust of these papers is that firms with substantial

debt facing severe financial constraints lay off workers more frequently. The capital-labor decision

is not the focus of this work.

Michelacci and Quadrini (2005) relate financial constraints to employee wages and claim, as we

do, that employees can be a good source of financing for constrained firms. They do not, however,

study the informational advantage of labor over physical capital, and we show that financially

constrained firms will grow more slowly, while Michelacci and Quadrini argue the opposite. In our

empirical work we present an explicit measure of financial constraints and find evidence consistent

with our theory.

Peek and Rosengren (2000), Cetorelli and Gambera (2001), Klein, Peek, and Rosengren (2002)

and Burgess and Pande (2003), in work on financial market imperfections, have emphasized the

broad damage that poorly functioning national or local credit markets can cause to investment

and growth. We study how financial constraints can distort one particular investment decision, the

capital-labor choice, and, extending this approach, we provide a theoretical analysis of the costs of

financial constraints.

Eisfeldt and Rampini (2005a,b) show that constraints can influence the type of capital used by a

firm. These papers demonstrate that constrained firms exhibit preferences for used capital (versus

new) and for leasing (rather than buying). Several previous studies have explored the effects of

the financing environment on the success of entrepreneurial firms and evaluated the comparative

advantages of bank and venture capital finance (Black and Strahan, 2002, Black and Gilson, 1998,

Ueda, 2004 and Inderst and Muller, 2004) These studies consider aspects of the capital choices

made by firms. Our central focus is on the optimal labor/capital strategies for firms to undertake,

given that they find it difficult to obtain credit.

3

Our paper complements other work that considers both the labor and the capital strategies of

firms. Several studies show that production in small firms is typically less capital intensive (Oi

(1983), Kimura (2001) and Dupuy and de Grip (2003)). Small firms are more likely constrained

than their larger counterparts, so these findings relate to our argument that financially restricted

firms employ less capital. Our empirical work, however, studies the effects of financial constraints

on investment and growth, with the analysis controlling for firm size.

2 Theoretical Model

2.1 Model Setup

We begin by describing the setup of the model.

2.1.1 Production in a Firm

We model an entrepreneur with a new business idea. Production takes place over two periods,

and it requires either labor or physical capital, which we will assume for simplicity to be perfect

substitutes. The entrepreneur chooses the mode of production (labor or capital) prior to the

beginning of each period. The idea of the entrepreneur and the input are then used to generate

output.

The productiveness of the firm (as determined by the quality of the entrepreneur’s idea or

management skill) is given by f . If labor is hired, the quality of the match between the labor and

the firm is q. If capital is hired or purchased, the quality of the match between the capital and

the firm is r. For convenience we assume that f , q and r are positive random variables with full

support over some bounded intervals. Total firm output π in a period is given by

π(f, q, r) ={

f + q if labor is hiredf + r if capital is hired

(1)

The entrepreneur must secure an input supplier for the second period before the first period

production is complete. At the conclusion of second period production, the business is closed down

and all assets have no further value. For simplicity, we set the discount rate equal to zero.

2.1.2 The Informational Advantage of Labor

The key distinction between labor and capital in the model is that the supplier of labor (i.e. a

worker) is capable of learning about the firm in the course of production, while the supplier of capital

(e.g. someone leasing a piece of machinery to the entrepreneur) does not receive information until

4

production is complete. Before the first period, the entrepreneur and the suppliers of labor and

capital are equally ignorant of their specific qualities. In the course of the first period production,

prior to the point at which second period arrangements must be made, however, f and either q

or r (depending on whether labor or capital is hired) are revealed to the entrepreneur. If labor is

hired, the worker also views f and q before making a decision about the second period. If, however,

capital is hired, the supplier of capital only views f and r when first period production is complete,

after second period arrangements have been finalized.

2.1.3 Inputs: Hiring or Purchasing

We assume for realism that labor may be hired, but capital may be either purchased or rented.3

Hiring (or renting) the input differs from purchasing it in two ways. First, if the input is hired,

either party may terminate the relationship at the end of the first period. Purchased capital, on

the other hand, belongs to the entrepreneur for as long as he wishes. Second, if the capital is

purchased, all output belongs to the entrepreneur. If, on the contrary, the labor or capital input

is hired, the supplier of the input acquires some bargaining power over the entrepreneur; we will

assume that the output is not verifiable, so contracting is not possible and bargaining takes place

over the division of the non-verifiable output.4

Since both the input and the entrepreneur are necessary for production, we will assume, as is

standard in the incomplete contracting literature (e.g. Hart and Moore, 1994), that the entrepreneur

receives a fraction 1 − θ of the output, where θ ∈ (0, 1), with the remainder going to the input

supplier (Binmore, Rubinstein and Wolinsky, 1986). Both labor and capital have opportunity costs,

which we denote by ε > 0 for labor and ζ > 0 for capital. Any labor or capital not employed by

the entrepreneur in a given period receives its opportunity cost as payoff that period. Capital may

be purchased or sold for 2ζ before the first period and for ζ before the second period.

We assume that the entrepreneurial venture is risky and, on average, not profitable5 for the

suppliers of the inputs (see, for example, Hamilton (2000), Moskowitz and Vissing-Jorgensen (2002)

and Audretsch (1991)). We assume that

θ(f + q

)< ε (2)

3None of the results change if capital can only be rented.4The robustness of the results to complete contracting is discussed in Section 3.3.4.5In our risk-neutral model we assume that entrepreneurial ventures are not profitable on average, but in a model

with risk-averse agents it would be enough to assume that they are insufficiently profitable to compensate agents fortheir substantial risk.

5

and

θ(f + r

)< ζ, (3)

where we denote the mean of a variable x by x.

2.1.4 Second-period Input Supply

If the relationship between the entrepreneur and the input supplier is terminated in the first period,

the entrepreneur may seek a new input for second period production. All outsiders (those who have

not participated in the firm) observe the age of the firm and whether any separations have occurred.

Outsiders, however, cannot distinguish between separations initiated by the entrepreneur and those

initiated by the input supplier (Lane, Isaac and Stevens (1996)).

2.1.5 Constrained and Unconstrained Entrepreneurs

We distinguish between two types of entrepreneurs. Financially constrained entrepreneurs have no

wealth with which to pay suppliers of capital or labor. Financially unconstrained entrepreneurs

have limitless wealth at their disposal and may offer advance payments to input suppliers. We

assume that the financial status of an entrepreneur is public knowledge.6

The timing of the model is summarized in Figure 1.

2.2 Results: The Capital-Labor Decision and Financial Constraints

We now consider differences in the capital-labor decisions of constrained and unconstrained en-

trepreneurs. We seek Perfect Bayesian Equilibria of the game described above. (All subsequent

references to equilibria are to Perfect Bayesian equilibria.)

Result 1. There is no equilibrium in which financially constrained entrepreneurs hire or buy

capital. A necessary condition for the existence of an equilibrium in which financially constrained

entrepreneurs hire labor is

θ(f + q) + E[max{θ(q + f), ε}] ≥ 2ε. (4)

A proof of Result 1 is given in the Appendix.6Several determinants of financial constraints such as local banking market concentration and entrepreneur eth-

nicity may be easily observed. Others are less easily observed, and we discuss the robustness of our results to thisassumption in Section 3.3.3.

6

It is clear that the purchase of capital is outside the means of the constrained entrepreneur,

but Result 1 shows that constrained firms cannot rent capital either. The intuition for the result is

as follows. Suppose that a constrained entrepreneur could rent capital in the first period. Capital

suppliers do not observe firm quality, so if a firm separates from its first-period capital supplier it is

deemed to be an average quality firm by the market (i.e., the separation conveys no information).

Assumption (3), that the one-period average returns to capital are less than the reservation rental

price, implies that no supplier will be willing to offer new capital to a firm in the second period

without an advance payment, which the constrained firm cannot offer. As a result, constrained

entrepreneurs who rent capital in the first period will always retain the capital for both periods.

Assumption (3) then shows that first-period suppliers of capital will not earn their reservation

payoffs over the two periods. It must therefore be that capital can only be rented with an advance

charge, which implies that constrained entrepreneurs cannot rent capital in equilibrium.

Why can constrained firms hire labor? Labor differs from capital in that the first-period em-

ployees can observe the firm’s quality f and the quality q of their labor match before making

second-period arrangements. If the sum of the firm and labor match qualities is too low, then the

employee will leave the firm. This real option to only remain with a successful venture enhances the

two-period payoff of labor hired in the first period from 2θ(f + q) to θ(f + q)+E[max{θ(q +f), ε}],and if (4) holds, then constrained firms can hire labor with no initial signing payment. The basic

intuition is that a worker may be willing to essentially finance the firm by accepting low initial

wages, because if he learns that the entrepreneur is of poor quality then he can simply leave. The

ability of workers to learn about a firm’s prospects and hence remain only with successful ventures

generates a real option that enables them to join risky new businesses, even without an initial

compensatory payment. Suppliers of capital do not learn as quickly as workers do, and hence can

only rent out their assets to firms with the ability to offer some fee in advance.

Result 2 shows that unconstrained entrepreneurs may hire capital.

Result 2. Unconstrained entrepreneurs may produce output in equilibrium with either labor or

capital. The unconstrained entrepreneur’s choice of input is determined by the relative productivities

and costs of labor and capital.

A proof of Result 2 is given in the Appendix.

Unconstrained entrepreneurs have sufficient cash to simply buy physical capital when that is

the input they prefer. Unconstrained entrepreneurs do not need financing, so their choice of input

7

is thus solely governed by the productivities and costs of labor and capital, without reference to

the fact that labor may be rented without an advance payment.

While in practice the differences between the two entrepreneurs will not be quite as stark as

depicted in the theoretical findings, the clear empirical implication of Results 1 and 2 is that

constrained firms will rely more heavily on labor while unconstrained firms will make more use of

capital. We will test this prediction in Section 4.

3 Model Extension: Firm-Specific Investments and Growth

3.1 Model Extension Setup

We now introduce a more comprehensive model of labor production that allows for an analysis of

firm-specific investment by the worker. As our focus in this section will be on the strategy of the

worker, we will analyze the case in which both constrained and unconstrained entrepreneurs hire

labor, not capital. (It is sufficient, for example, to assume that the conditions given in the proofs

of Results 1 and 2 for labor-hiring equilibria hold.)

Each period, after having viewed the firm and labor match qualities, the worker decides whether

or not to make a firm-specific investment. This investment requires some effort by the worker, in

exchange for which he becomes more productive within the firm. For example, the employee may

exert himself to master the firm’s internal systems. This skill enables the employee to produce

more while working with the entrepreneur but does not increase his value to other employers.

We presume that if the employee exerts effort, he pays a personal cost of g, and the quality of

his labor match permanently improves from q to q + λ. If the worker fails to exert effort, no gains

are realized. We assume that firm-specific investment is essentially a long-term investment; it is

only profitable to the worker when he expects to enjoy the benefits over both periods. Specifically,

g > θλ > 0 (5)

and

2θλ > g. (6)

The timing of the extended model (for the case of the labor-hiring equilibrium) is summarized

in Figure 2.

8

3.2 Results: Firm-specific Investment, Growth and Financial Constraints

We now describe the equilibrium firm-specific investment decisions of workers in constrained and

unconstrained firms and analyze the impact of these choices on firm growth.

The second period firm-specific investment decision of the worker is straightforward. Inequality

(5) shows that the optimal second-period strategy for workers at both constrained and uncon-

strained firms is to not exert effort.

In the first period, a worker will be willing to make a firm-specific investment only if he expects

to be retained in the second period, and if he expects to reap some benefit from the investment in

the second period. Lemma 1 gives a condition under which a constrained entrepreneur will retain

a worker.

Lemma 1. If worker quality q at the end of the first period satisfies

θ(f + q) ≥ ε, (7)

then the worker will be retained by a constrained entrepreneur.

A proof of Lemma 1 is given in the Appendix.

The following is the intuition for Lemma 1. If a constrained firm seeks a new worker after the

first period, this is regarded as a weakly negative signal about the firm by the market, because

the first-period separation may have arisen from a firing or a quitting (or both), and none of

these indicates good news about the firm. Condition (2), that even average firms insufficiently

compensate employees, thus guarantees that no new worker will join the constrained firm without

an advance payment, which the firm cannot provide. As a consequence, whenever the firm is able

to retain its first period employee (i.e., whenever (7) holds), it will do so. This generates an implicit

retention guarantee for employees of successful constrained firms.

We emphasize that the overall probability of retention need not be higher in constrained firms.

Constrained firms may be forced, due to lack of resources or complete firm failure, to lay off their

employees. When (7) fails, only unconstrained entrepreneurs can retain their employees. The

essential point is that when the firm is successful (i.e., when (7) holds), constrained firms are more

likely to retain their employees than unconstrained firms, but the opposite is true when the firm is

struggling.

Result 3. If a worker in an unconstrained firm makes a firm-specific investment, then a worker

9

in a constrained firm makes a firm-specific investment. The converse does not hold.

Retention alone is not sufficient to induce the worker to exert effort; he must also realize a

benefit from his effort to improve his productivity. If (7) fails, the worker may quit, be fired or

offered a retention payment to stay, but in all cases he will simply receive a net payment of his

reservation value ε. Employees do not benefit from being retained by poorly-performing firms,

and the prospect of being retained by such a firm will not encourage them to make a firm-specific

investment. If (7) holds, the worker will receive a share of the output that he has increased. It

is thus an implication of Lemma 1 that constrained firms retain workers precisely when it is in

the interest of the workers to make firm-specific investments, and workers understand this. The

practice of retention by successful constrained firms is the reason that entrepreneur and worker

incentives are better aligned in constrained, rather than unconstrained firms, despite the greater

survival probability of the latter. As a consequence, Result 3 predicts that workers should be more

willing to invest in skills specific to the firm when working for financially restricted entrepreneurs.

Result 3 shows that successful constrained entrepreneurs induce employee investment in their

firms by offering implicit retention guarantees to workers. Constrained entrepreneurs, however,

also suffer from the fact that they cannot dismiss poor quality workers. This argument suggests

that constrained entrepreneurs will have relatively high first-period output (due to firm-specific

investments by workers) but relatively low second-period output (because they have lower quality

workers). Result 4 formalizes this idea. We denote the constrained firm’s output in period i = 1, 2

by πi,c, and πi,u is defined similarly for unconstrained firms.

Result 4. Constrained firms experience relatively high first-period output and relatively low

second-period output:

E[π1,c

] ≥ E[π1,u

]

and

E[π2,c

] ≤ E[π2,u

].

Moreover, conditional on first-period output, unconstrained firms have higher expected second-

period output than constrained firms. Formally, for all x ≥ 0,

E[π2,u|π1,u = x

] ≥ E[π2,c|π1,c = x

].

10

Result 4 states that constrained firms will have lower output growth than unconstrained firms.

In the model, the firm’s assets and number of employees do not vary over time, so Result 4 essentially

predicts lower labor productivity growth for constrained firms. This result complements some of

the findings in the literature that link financial constraints to slower growth due to reduced physical

capital investment (Demirguc-Kunt and Maksimovic, 1998, Carlin and Mayer, 2003 and Almeida

and Campello, 2005). In contrast, Result 4 focuses on human capital: constrained firms benefit

from greater firm-specific investment by their workers, but gradually suffer from a negative selection

effect in the quality of their employees that arises from their limited ability to fire. Result 4 predicts

that constrained firms will perform relatively well early in their lives and relatively poorly later for

reasons that are completely independent of their physical capital investment policies.7

3.3 Robustness

We now consider the effects on the theory’s implications of relaxing several modelling assumptions.

3.3.1 Correlation between firm quality and financial status

One possible extension to the model would be to allow for positive correlation between firm quality

f and unconstrained status. That is, better entrepreneurs may be wealthier. In this case, ability

to pay input suppliers in the second period would now signal firm quality f and the relative

advantage of unconstrained entrepreneurs in second period hiring described in the base model

would be enhanced. Since this relative hiring advantage drives the results, the main findings of the

paper would continue to hold.

3.3.2 First-period payoffs realized before inputs secured for second period

A second possible extension would be to have first-period payoffs realized by the entrepreneur

before second period arrangements are made. In such a setting, some very successful constrained

firms would switch to unconstrained status, which would moderate the distinction between initially

constrained and unconstrained entrepreneurs. Nonetheless, ability to pay in the second period

would signal firm quality in this extension as well, and the relative hiring advantage of unconstrained7Result 4 leaves open the question of the overall impact of financial constraints on total output. Can the benefits of

firm-specific investment outweigh the costs of retaining poor quality workers? An example provided in a previous draftshowed that if the worker is of moderately less than average quality and the gains from firm-specific investment arelarge, then constrained entrepreneurs can have a higher two-period expected payoff than unconstrained entrepreneurs.The thrust of this example is not to suggest that entrepreneurs are better off if they are financially constrained, butrather to argue that the relative benefits of being financially unrestricted are smallest when current employees areonly slightly sub-par and firm-specific investment yields large gains.

11

entrepreneurs would be maintained. Thus, the central implications of the paper are robust to this

extension as well.

3.3.3 Financial status is private information

A third extension would require that financial status be unknown to potential investors and em-

ployees. A pooling equilibrium in which all entrepreneurs rent capital would require that no initial

payments for capital be made (because the constrained cannot make these payments), but the

argument underlying Result 1 indicates that such an equilibrium cannot exist. As a result, con-

strained entrepreneurs would be unable to rent capital in this model as well, and Results 1 and 2

would continue to hold. Results 3 and 4, however, do require that employees be informed about the

financial status of the entrepreneur when making their firm-specific investment decisions. Results

3 and 4 thus require that employees be able to learn about the entrepreneur’s financial condition

in the course of employment. Results 1 and 2 are robust to this assumption, as well.

3.3.4 Complete contracts

A final variation on the base model would be to allow for full verifiability of firm cash flows and

complete contracts. For example, the employee could be offered a fixed or contingent wage, and the

physical capital supplier might be awarded a debt or equity claim. This variation would have little

effect on the capital-labor decision analyzed in Section 2. The central issues in that section are the

information disadvantage of capital relative to labor and the desire of the entrepreneur to continue

the project even if it not profitable for the input supplier. An unconstrained entrepreneur can always

purchase capital, while a constrained entrepreneur will be unable to rent capital, irrespective of the

contract he offers, because the project is on average not worthwhile. A worker, by contrast, may

join a constrained firm because of his ability to leave if project quality is revealed to be low.

The verifiability of cash flows would, however, affect Results 3 and 4. The intuition for these

results is that financially constrained status essentially serves as a commitment device on the part

of the entrepreneur to retain the worker because successful constrained entrepreneurs do not replace

their employees, while successful unconstrained entrepreneurs may do so. This implicit commitment

to retain their employees induces them to invest in the firm in the first period. In the presence of

verifiable cash flows, however, contracts may be used by unconstrained entrepreneurs to make an

explicit retention promise. This suggests that the model of firm-specific investment described in

Section 3 is most appropriate for firms in which cash flows are quite costly or hard to verify. This

condition is likely to apply to many entrepreneurial ventures for which the use of external cash flow

12

monitors is simply impractical.8

4 Empirical Tests: Capital-Labor Ratios, Productivity Growthand Financial Constraints

In this section we provide empirical tests of Results 1, 2 and 4. Section 5 provides tests of Result

3.

4.1 Data

Our data source for these tests is the 1998 National Survey of Small Business Finance (NSSBF)

conducted by the Board of Governors of the Federal Reserve System. This survey, which has

been used by a number of researchers (e.g. Moskowitz and Vissing-Jorgensen (2002) and Petersen

and Rajan (2002)) collects information on owner demographics, financing characteristics, balance

sheets and other attributes of small businesses (which are defined to be firms with fewer than 500

employees).

There are 3,561 U.S. firms in the data set. Firms report the book value of assets and the number

of employees they hire. The value of assets includes capital that is rented in the form of capitalized

leases or secured with collateralized loans but excludes operating leases. Data are also provided on

sales revenue, the total cost of conducting business, profit margin (i.e., profits divided by sales),

firm age, the Dun and Bradstreet firm credit score (a 1-5 scale) and whether the firm is located

in an MSA (Metropolitan Statistical Area). A bank competition index from one to three is given

that describes the commercial bank deposit Herfindahl measure of the firm’s MSA (if it is in an

MSA) or county. (The firm’s location, however, is not revealed.) The firms also report financial

information including whether their most recent loan application was rejected and whether they

have applied for a loan in the last three years.

Detailed data on the owner is provided, such as the value of home equity, net wealth (excluding

both home equity and the value of the firm), ethnicity, gender and education (on a 1-7 scale). The

52 firms with zero sales, the 7 firms with negative asset levels and the 1 firm without a reported

bank Herfindahl index are excluded from the data set. Summary statistics including means (all

variables) and medians and standard deviations (all non-binary variables) are given in Table 1.8The assumption of privately observed cash flows is found in many settings, including Harris and Raviv (1995),

Gromb (1999) and DeMarzo and Fishman (2002).

13

4.2 Empirical Tests: The Capital-Labor Decision and Financial Constraints

4.2.1 Theoretical Hypothesis and Econometric Strategy

Results 1 and 2 suggest that constrained entrepreneurs will make relatively greater use of labor

in production while unconstrained entrepreneurs will rely more heavily on capital. To test this

hypothesis, we estimate the following equation:

capital to labor ratioi = α + β ∗ (financially constrainedi) + γ ∗ controlsi + εi, (8)

where capital to labor ratioi is set equal to log(1 + value of assetsi

number of employeesi

), financially constrainedi

is a proxy for the probability that an entrepreneur is constrained, controlsi is a vector of controls

containing firm and owner attributes and εi is an error term. The implication of Results 1 and 2 is

that β < 0: financially constrained firms have lower capital-labor ratios.

The NSSBF data set of small firms is appropriate for testing the theory because the issues

highlighted in the model, in particular individual employee contributions that have a significant

effect on firm-wide outcomes, the large risk of bankruptcy and asymmetric information about the

firm’s financial status and quality, are most germane in smaller firms.

4.2.2 Instruments for Financial Constraints

Estimating (8) requires a reasonable proxy for the probability financially constrainedi that a

given firm i is financially constrained. A clear indication of financially constrained status is

that a firm’s loan application is denied, and we use loan application denial as our measure of

financially constrained. Loan application denials are only informative about firms that make a

loan application, so we restrict our data to the 952 firms that made applications (and that meet

the data requirements described earlier).

It is reasonable to consider that firms with small capital-labor ratios may have very few assets

with which to secure a loan, and that this lack of collateral may cause their loan applications

to be denied, rather than the converse. That is, the relationship between a loan denial (i.e.,

financially constrained) and a low capital-labor ratio may be endogenous. We will therefore

provide instruments for loan denial and use these instruments as measures of financial constraints.9

We propose the following eight instruments for whether a firm has its loan application denied:

firm presence in a concentrated local banking market, the entrepreneur’s home equity and other9In addition, it may be the prospect of having a future loan request denied, rather than the actual denial itself,

that weighs heaviest on most constrained firms.

14

wealth, the entrepreneur’s education level, gender and African-American or Hispanic ethnicity and

the firm’s Dun and Bradstreet credit score. The rationale for these instruments is as follows. Con-

centrated banking markets, small entrepreneurial wealth (and hence little collateral), low education

level, female and minority status and poor firm credit ratings are all variables that may plausibly

make it more likely that an entrepreneurial applicant will be denied a loan. On the other hand, we

presume that these variables will not have a direct effect on firm capital-labor ratios.10 We assume,

for example, that in the absence of financial constraints, African-American entrepreneurs would

not choose capital-labor ratios that differ systematically from those of other entrepreneurs.

In the main model we presumed that the entrepreneur’s financial status was public information,

and that is a reasonable description of the following instruments: local banking market concen-

tration, entrepreneur education, gender and ethnicity, and firm credit score. The entrepreneur’s

home equity and other wealth are less easily observed. In Section 3.3.3 we argue that the results in

the model are robust to assuming that workers can learn the entrepreneur’s financial status after

being employed for some time. Therefore, all we need to assume is that workers can learn about

the entrepreneur’s home equity and other wealth in the course of their employment.11

Our central strategy will be to estimate a 2SLS (two-stage least squares) regression in which

the first stage equation is given by

financially constrainedi = F (instrumentsi, controlsi, νi), (9)

where financially constrainedi is a binary variable set equal to one if the firm was denied a loan,

F is a linear function, instrumentsi is a vector of the eight instruments described above, controlsi

is a vector of controls containing firm and owner attributes and νi is an error term. The second

stage equation is (8). That is, even though our first stage dependent variable is binary, we estimate

2SLS using a linear form in the first state, as recommended by Angrist (2000). We estimate (8)

and (9) via a cross-sectional 2SLS regression that assumes that errors are independent across firms

but potentially heteroskedastic.

To simply evaluate the appropriateness of the instruments, without estimating the full 2SLS,

we also consider a logistic form for F , which reflects the binary nature of financially constrained.10In the model, unconstrained entrepreneurs have limitless cash, so even if, for example, the cost of borrowing is

higher in more concentrated banking markets, unconstrained entrepreneurs can simply buy physical capital outrightand bypass the banking market. As a result, local banking market competition will not have an effect on thecapital-labor ratio of unconstrained entrepreneurs.

11Workers will learn where the owner lives, what car he drives, where he vacations, etc. As Section 3.3.3 makesclear, Results 1 and 2 are also robust to assuming that financial status is unobservable, but Results 3 and 4 requirethat workers know something about the firm’s financial status after having been employed for a period.

15

A second potential proxy for financially constrained that we do not utilize is an indicator for

whether or not the firm made any loan applications in the last three years. We do not make use of

this proxy because not applying for a loan may indicate either constrained status or unconstrained

status or neither:

• Constrained firms may be so unlikely to receive financing that they are discouraged from

applying. In this case, not having applied for a loan would be associated with constrained

status.

• Unconstrained firms may have sufficient internal capital or long-term loans such that they

need not apply for new debt. In this case, it is the unconstrained firms that do not apply.

• It may also be that filing a loan application is relatively low cost, so that it is done by both

constrained and unconstrained firms.

4.2.3 Testing the Instruments

We begin by evaluating whether the instruments are associated with our measure of financial

constraints. In the first column of Table 2 we report results from regressing a dummy variable

for whether the firm’s most recent loan application was rejected on the instruments and a set of

controls. The controls with reported coefficients are log of one plus sales, log of one plus firm age,

profit margin and an indicator for firm location in an MSA. Additional controls with coefficients

unreported for brevity include a constant, industry dummies at the 2-digit SIC level, census region

dummies, the log of one plus the owner’s experience in managing a business, the log of the owner’s

age, the ownership share of the primary owner, a dummy for whether the firm is managed by the

owner, a dummy for firms with zero sales in the previous year, a dummy for new firms, a dummy for

firms subject to corporate income tax and a dummy for family-owned firms. The estimation is via

binary logistic regression (Logit), with t-statistics reported in parentheses using robust “sandwich”

(White) standard errors.

We find that several instruments enter equation (9) with statistically significant coefficients.

Firms are significantly (t = 2.36) more likely to be denied a loan in concentrated local banking

markets.12 Owners with less home equity (t = −3.08) and other wealth (t = −2.07), African

Americans (t = 5.58) and Hispanics (t = 2.98) are all significantly more likely to have their12This finding is consistent with the evidence in Hannan (1997) and Berger, Rosen, and Udell (2001) that concen-

trated markets are less competitive. Black and Strahan (2002) and Cetorelli and Strahan (2004) show that bankingconcentration has an especially deleterious effect on small firms.

16

loan applications rejected. Firms with better Dun and Bradstreet credit scores are also weakly

(t = −1.94) more likely to receive a loan. Entrepreneur education and gender do not have a

significant effect on loan denials. Taken together, however, the results in the first column of Table

2 support the use of the proposed instruments.

4.2.4 Empirical Findings: Capital-Labor Ratios and Financial Constraints

Results 1 and 2 imply that financially constrained firms will have low capital-labor ratios, and we

test this hypothesis by estimating (8) and (9) via 2SLS. In column two of Table 2 we report results

from the 2SLS regression of capital-labor ratios on whether the most recent loan was declined. We

make use of the eight instruments and all the controls previously described in Section 4.2.3. (The

NSSBF does not provide detailed firm locations or physical capital or labor market variables, so the

regressions do not include prices of physical capital or labor.) We find that financially constrained

firms (i.e., those whose loan applications were declined) have substantially reduced capital-labor

ratios, as predicted by Results 1 and 2. The t-statistic on loan denial is -3.93 (calculated with

robust standard errors).13 For a given firm, an increase from zero to one in its probability of having

a loan declined decreases its capital-labor ratio by 23.6 percent of the mean.

Given that we have eight instruments and only one endogenous variable, we conduct a heteroskedasticity-

robust test of overidentifying restrictions (Wooldridge, 2002, p.123). This test assesses whether the

proposed instruments are actually exogenous. We find a p-value of 0.41, suggesting no reason to

reject the validity of the instruments.

In the third column of Table 2 we report results from repeating the previous regression, but

making use of only the following three instruments: local banking market concentration, the log of

the owner’s home equity and the log of the owner’s net worth, because these are perhaps the most

closely linked to the idea of financial constraints described in the theoretical model. The t-statistic

on loan denial is -3.15 in this specification, and the p-value for the overidentifying restrictions is

0.68.

The idea underlying Results 1 and 2 that employees are quicker to learn firm quality than

suppliers of physical capital is likely most important for young firms. For older firms, input suppliers

face less of an asymmetric information problem, so learning may be less important. To test this

idea, we create subsamples of young (firm age less than or equal to 11) and old (firm age above13A simple direct OLS (ordinary least squares) regression of capital-labor ratios on whether the most recent loan

application was denied also yields a negative and significant coefficient, but we find that the 2SLS results offer theclearest interpretation.

17

11) firms and repeat the eight-instrument 2SLS regression in each subsample. (The median firm

age is 11.) The results, displayed in columns 4 and 5 of Table 2 show that the coefficient on loan

denial is significant (t=-3.58) in the young firms subsample but insignificant (t=-1.22) in the old

firms subsample. This is consistent with the argument of the model that constrained firms’ lower

capital-labor ratios are driven by the informational advantage of labor.

4.2.5 Robustness Tests: The Capital-Labor Ratio and Financial Constraints

As a robustness check for the results described in Section 4.2.4, we estimate (8) via OLS using the

instruments as proxies for financially constrained. This is not a formal 2SLS test, but it provides

information on the effects of the different instruments on capital-labor ratios. If an instrument for

financially constrained status was associated with higher capital-labor ratios, this would be a cause

for concern. In the sixth column of Table 2 we report results from regressing capital-labor ratios on

the eight instruments directly, and we include all the previous controls. Firms whose owners have

more home equity (t = 2.70) and net wealth (t = 2.01) and firms with high credit scores (t = 1.79)

have higher capital-labor ratios, while firms in banking markets with high Herfindahls (t = −1.69)

and firms with African-American owners (t = −2.15) or female owners (t = −1.72) have lower

capital-labor ratios. The coefficients on education and Hispanic ethnicity are insignificant. The

results depicted in the first column of Table 2 show that lower home equity, lower net wealth, lower

firm credit scores, higher banking market concentration and African-American and Hispanic status

of owners and are all associated with financial constraints (loan denials), and we find that the first

five of these variables are also linked in a statistically significant manner to lower capital-labor

ratios, while the sixth exhibits an insignificant correlation. Overall, this provides broad support for

the argument that the empirical effect of financial constraints on capital-labor ratios is consistent

across instruments.

As an additional test, we drop all observations for which the capital-labor ratio is above its 95th

percentile or below its 5th percentile and find that this screening has essentially no effect on any

of the regressions in Table 2.14

4.2.6 Robustness Tests: The Labor Intensity of Production and Financial Constraints

Results 1 and 2 have a direct implication for the relative capital-labor ratios of constrained firms,

but they also have the more general implication that financially restricted firms will produce in14For example, the coefficient on financial constraints in the 2SLS regression described in column 2 of Table 2

decreases from -2.344 to -2.3473 and the t-statistic decreases from -3.93 to -4.01.

18

a labor-intensive manner. In the model, the total cost of conducting business is the payment to

the input supplier, so the theory predicts that the costs per employee will be lower in constrained

firms as they hire more workers and use less physical capital. To test this broader implication, we

relate the total costs of conducting business per employee to a firm’s financial status. Specifically,

we estimate the following equation:

yi = Φ + ∆ ∗ (financially constrainedi) + κ ∗ controlsi + Σi, (10)

where yi is set equal to log(1 + total costsi

number of employeesi

), financially constrainedi and controlsi are

defined as in (8) and Σi is an error term. The total cost of conducting business includes rental

costs on production facilities, the costs of all materials and interest payments on purchased capital,

so this specification allows for the possibility that constrained firms may make greater use of

operating leases or purchase (rather than produce) materials. All these costs are incorporated

in log(1 + total costsi

number of employeesi

), which is a measure of the labor intensity of production that thus

controls for varying leasing or outsourcing strategies by different firms. (The NSSBF does not

provide data on the total wage bill of the firm.) The implication of Results 1 and 2 is that ∆ < 0:

financially constrained firms have lower total costs-labor ratios. We jointly estimate (9) and (10)

using 2SLS.

In column one of Table 3 we report results from the 2SLS regression of total cost-labor ratios

on our measure of financial constraints (i.e., whether the most recent loan was declined). We make

use of the eight instruments and all the controls previously described in Section 4.2.3, with the

exception of the profit margin, since the total costs are used to compute the profit margin. The

four firms with zero costs are excluded. We find that financially constrained firms have significantly

(t=-3.68) lower total cost-labor ratios, as predicted by Results 1 and 2. For a given firm, an increase

from zero to one in its probability of having a loan declined decreases its total costs-labor ratio

by 9.6 percent of the mean. The p-value for the overidentifying restrictions is 0.80. Column two

of Table 3 details the results from the running the same regression but making use of only local

banking market concentration, the log of the owner’s home equity and the log of the owner’s net

worth as the instruments in Equation (9). We find similar results, with a negative and significant

coefficient on financial constraints (t=-2.16) and an overidentifying restrictions p-value of 0.21.

We next consider the relationship between a firm’s financial status, the number of its employees

and its depreciable, depletable and intangible assets. (For simplicity, we refer to the latter class

of assets, which are treated as a group in the NSSBF, simply as the firm’s depreciable assets.)

19

As a second test of the effects of financial constraints on the labor-intensity of production, we

estimate (10), setting yi equal to log(1 + depreciable assetsi

number of employeesi

). Depreciable assets may be more

likely substitutes for labor than other assets such as land or cash. As above, the implication of

Results 1 and 2 is that ∆ < 0. We jointly estimate (9) and (10) using 2SLS.

In column three of Table 3 we report results from the 2SLS regression of depreciable asset-labor

ratios on our measure of financial constraints. We utilize the eight instruments and all the controls

previously described in Section 4.2.3. The one firm with negative depreciable assets is excluded.

We find that financially constrained firms have significantly (t=-2.60) lower depreciable assets-labor

ratios, as predicted by the model. For a given firm, an increase from zero to one in its probability

of having a loan declined decreases its total depreciable assets-labor ratio by 38.1 percent of the

mean. The p-value for the overidentifying restrictions is 0.26. As shown in column four of Table

3, the results from the regression using only local banking market concentration, the log of the

owner’s home equity and the log of the owner’s net worth are again consistent with the previous

finding. We find a t-statistic of -2.37 on our measure of financial constraints, and the overidentifying

restrictions p-value is 0.21.

The empirical findings in Tables 2 and 3 thus together provide a variety of evidence in favor of

the prediction of Results 1 and 2 linking financial constraints to lower capital-labor ratios and to

greater labor-intensity of production, more broadly.

4.3 Empirical Tests: Firm Productivity Growth and Financial Constraints

4.3.1 Theoretical Hypothesis and Econometric Strategy

Result 4 predicts that constrained firms will experience slower productivity growth than uncon-

strained firms, controlling for the level of physical capital and number of employees in the firm. To

test this prediction, we estimate the following productivity equation:

log(salesi) = (11)

ψ1 + ψ2 ∗ (financially constrainedi) + ρ ∗ (financially constrainedi) ∗ log(1 + firm agei)

+ψ3 ∗ log(1 + firm agei) + ψ4 ∗ (firm productivity controlsi) + ui

20

where salesi is the dollar value of firm i’s sales, firm agei is the age of firm i in years, firm

productivity controlsi is an augmented vector of firm and entrepreneur attributes and ui is an

error term. The implication of Result 4 is that ρ < 0: the relative productivity of constrained firms

decreases over time.

We include in firm productivity controls both the log of one plus assets and the log of employees

interacted with industry dummies, and measures of firm assets in different categories (e.g. cash,

land, inventory, etc.) This approach presumes that factor shares are constant within industries

(Kahn and Lim, 1998), and we control for the use of different asset types by different firms. Our

accounting for asset levels and types helps to ensure that our productivity results are not driven

by different levels of capital investment in constrained firms.

The results in columns four and five of Table 2 indicate that the learning problems faced by

input suppliers that are at the heart of the theoretical model are, as might be expected, most

salient for young firms. This suggests that the decreasing relative productivity of constrained firms

is likeliest to be found in the sample of younger firms. As firms age, information asymmetries

become less relevant, the assumptions underlying the model will gradually fail to obtain and the

difference between the productivities of constrained and unconstrained firms will begin to narrow.

Consequently, we separate the samples of old and young firm in our productivity analysis.

Ideally, we would test Result 4 by utilizing a panel data set allowing time-varying productivity

estimates for given firms over time, but the NSSBF data does not provide the historical data

required for that strategy. Instead, as described in (11), we estimate productivity for a cross-

section of firms and analyze the interaction between financially constrained status and firm age.

4.3.2 Labor Productivity versus Total Factor Productivity

Strictly speaking, Result 4 concerns labor productivity, not general productivity, so we interpret the

coefficients on financially constrained status and its interaction with firm age as labor productivity

effects of these variables. This follows the literature on labor productivity (e.g. Freeman and

Kleiner, 2000 and Black and Lynch, 2001). As in any Cobb-Douglas type specification, however, in

(11) we are actually estimating total factor productivity, so it might be argued that any productivity

differences we find between financially constrained and unconstrained firms actually arise from

differences in asset productivity. Against this point we note that we measure assets in dollar terms

and control for the various types of assets used by the firm. It seems unlikely that the productivity

per dollar of assets varies widely within an industry, controlling for level of assets and asset type.

21

Moreover, the prediction of Result 4 is that the relative productivity of constrained firms will fall

over time. This interaction result would not be generated by any difference between the productivity

of assets in constrained and unconstrained firms that is constant over time.15 Result 4 is distinctive

in that it predicts a time series effect on the relative productivity of constrained firms.

4.3.3 Instrumental Variables

To estimate (11) we require a proxy for financially constrained. We do not make direct use of the

indicator for whether a firm’s loan application is denied, because this variable is endogenous is this

setting as well; it may be that older firms are only denied a loan if they are especially unproductive.

The analysis is somewhat complicated by the fact that both financially constrained status and the

interaction between constrained status and firm age are endogenous. That implies that we must

instrument for both variables using a form of non-linear 2SLS.

We adopt the approach suggested in Angrist (2000) and Kelejian (1971) for non-linear 2SLS.

Specifically, we first estimate

financially constrainedi = G(instrumentsi, firm productivity controlsi, ωi), (12)

where financially constrainedi is a binary variable set equal to one if the firm was denied a loan,

G takes the logistic form, instrumentsi is the vector of the eight original instruments described in

Section 4.2.2 and ωi is an error term. We denote the predicted value of financially constrained

generated by the estimation of (12) by f c.

We use this predicted variable to construct two new instruments ninstruments: f c and f c ∗log(1 + firm age). The variables ninstuments are used as instruments for financially constrained

status and the interaction between financially constrained status and the log of firm age. We use

ninstruments to perform 2SLS. The first stage equations are

[financially constrainedi

(financially constrainedi) ∗ log(1 + firm age)

]= Γ

constantninstrumentsi

log(1 + firm agei)firm productivity controlsi

+ vi,

(13)

where Γ is a matrix of coefficients to be estimated and vi is a vector of error terms. The second

stage equation is given by (11). We estimate (11) and (13) via a cross-sectional 2SLS that assumes15In the Eisfeldt and Rampini (2005a,b) models, for example, each firm produces in only one period, so a given

firm does not experience changing productivity over time.

22

independent but potentially heteroskedastic errors across firms.

4.3.4 Empirical Findings: Firm Productivity and Financial Constraints

We begin by analyzing the effects of financial constraints on productivity in the sample of older

firms. In the first column of Table 4 we report results from the regression of the log of sales on

financially constrained status, the interaction of financially constrained status and the log of one

plus the firm age, the log of one plus the firm age and the set of firm productivity controls. The

firm productivity controls include all the variables used in the regressions in Table 2 (detailed in

Section 4.2.3) and also include the log of one plus assets and the log of employees interacted with

industry dummies, and the log of one plus the asset type for the following eight asset categories:

cash, accounts receivable, inventory, other current assets, other investments, land, buildings plus

equipment and other assets. As an additional control, we use the geographic size of the firm’s

market (measured on a 1-5 scale); this is designed to capture productivity differences between

exporters and non-exporters. The estimation is via 2SLS, using the two new instruments generated

by the eight original instruments, as described in Section 4.3.3, with robust t-statistics.

The results in the first column of Table 4 show that in the sample of older firms, the estimated

coefficient on the interaction between financial constraints (i.e., loan denial) and firm age is neg-

ative but insignificant. In the second column of Table 4, we report results from conducting the

same regression on the sample of younger firms. For the set of younger firms, the coefficient on

the interaction is negative and significant (t=-2.12), as predicted by Result 4.16 The estimated

coefficients indicate that an increase from zero to one in the probability of having a loan declined

raises the log of sales by 0.7% of the mean for one-year-old firms, but lowers it by 8.0% of the mean

for eleven-year-old firms, controlling for the level of assets and number of employees in the firm. As

suggested by Result 4, within the group of firms for which information asymmetries are important,

financial constraints are a greater disadvantage as firms age.

We also conduct these regressions exploiting only local banking market concentration, the log of

the owner’s home equity and the log of the owner’s net worth as the instruments utilized in equation

(12) to generate the two new instruments ninstruments. The results, shown in columns three and

four of Table 4 confirm our previous findings. The interaction between financial constraints and

firm age is insignificant in the older firms sample, and negative and significant (t=-2.31) in the

younger firms sample. Dropping observations for which sales is above its 95th percentile or below16A direct OLS regression of log of sales on whether the most recent loan application was denied interacted with

firm age also yields a negative and significant coefficient, but we focus on the instrumented results.

23

its 5th percentile has only a minor effect on the coefficient estimates and does not change the

statistical inference for any of the regressions in Table 4.17

The results in Table 4 indicate that firm age has no significant effect on productivity for older

firms of either the constrained or unconstrained types. Unconstrained younger firms, however,

experience significant positive productivity improvements over time, while constrained younger

firms do not experience any significant change. This finding is consistent with the mechanism in

the model by which unconstrained firms can improve their productivity by firing poor employees

and hiring new ones who are better on average, while constrained firms retain their existing workers

and maintain a fairly stable level of productivity over time. The empirical results described in Table

4, particularly those for young entrepreneurial firms, thus provide support for Result 4.

5 Empirical Tests: Firm-Specific Investments and Financial Con-straints

In this section we provide tests of Result 3.

5.1 Data

Our data source for the tests of Result 3 is the 1996-1997 National Organizations Survey (Kalleberg,

Knoke and Marsden, 1996-1997). The National Organizations Survey (NOS) gathers data from U.S.

work establishments on their formal structures, internal labor markets and use of hierarchies. There

are 1002 organizations in the data set. We make use of two variables in the NOS on organizational

practices. The first is a dummy variable indicating whether employees in the firm are involved in

work teams when they do their job. The second is a summary measure of the average level at

which decision are made within the firm. To construct the second variable we make use of data

in NOS that describes who within the firm makes the final decision in the following eight areas:

new employee hiring, the use of subcontractors, the use of temporary workers, worker evaluation,

product or service improvement, worker schedules, production targets and the provision of training

programs. For each area the respondent indicates if the decision is made by the head of the

organization (to which we assign a score of four), a middle manager (three), a supervisor (two) or

someone below (one). For each firm we then average the scores across all the areas for which data

is available to generate the summary measure.17For example, removing sales outliers from the regression described in the second column of Table 4 increases the

coefficient on the interaction between financial constraints and log of firm age from -0.629 to -0.557 and increases thet-statistic from -2.12 to -2.11.

24

The NOS includes location data which enables us to calculate the commercial bank deposit

Herfindahl of the organization’s MSA (if it is in an MSA) or county (otherwise) in 1996. The owner

demographic and credit score variables we employed in Table 2 are not available in the NOS data.

We restrict attention to organizations that are small firms. Our requirements that organizations

have 500 or fewer employers, have for-profit status and provide the necessary organizational data

are met by 241 firms in the data set.

5.2 Theoretical Hypotheses and Econometric Strategy

Result 3 shows that a worker will be more willing to make a firm-specific investment in a constrained

firm than in an unconstrained firm. In order to generate empirical implications from Result 3, we

must make the idea of a firm-specific investment concrete. In a previous draft (available from

the author) we considered two examples of firm-specific investments: workers investing in co-

specialization with fellow employees and workers choosing projects that generated firm-wide benefits

at some personal cost. Our analysis of both these particular examples was driven by the intuition

underlying Result 3 that employees in constrained firms are more willing to make firm-specific

investments. We showed that constrained firms are more likely to produce in teams, because

workers in constrained firms are more willing to make the co-specialization efforts that make team

production efficient. We also showed that constrained entrepreneurs are more willing to grant their

employees discretion in project selection, as employees in these firms are more likely to choose

projects that generate firm-wide benefits, even at some personal cost.18

This suggests two empirical implications of Result 3; firms owned by constrained entrepreneurs

should more frequently produce in groups and should be more likely to grant their employees

significant autonomy (i.e., allow decisions to be made at a lower level). To test these implication,

we estimate the following equation:

zi = H(financially constrainedi, controlsi, ηi), (14)

where zi is a description of an organizational practice (production in teams or level at which

decisions are made), financially constrainedi is proxied for by the local commercial bank deposit

Herfindahl, controlsi is a set of firm-level controls and ηi is an error term. We estimate (14)

via cross-sectional Logit (for the binary team-production variable) and OLS (for the decision-level

variable), assuming that errors are independent across firms but potentially heteroskedastic.18Harris and Raviv (2002), Dessein and Santos (2003), Vayanos (2003) and Garicano (2002) provide other models

of the optimal form of firm organization.

25

5.3 Empirical Findings: Firm-Specific Investments and Financial Constraints

We analyze a logit regression of a dummy variable indicating whether employees in the firm are

involved in work teams on the concentration of the local banking market and controls for the

number of employees, sales, age and industry (at the 2-digit SIC level). Consistent with the first

implication of Result 3, we find that firms located in concentrated banking markets are significantly

more likely to have their core employees involved in work teams (t=2.59). A one-standard-deviation

increase in banking concentration increases the probability of work teams by 6.6% when the other

variables are evaluated at their medians.

We also regress via OLS the average level at which decisions are made on banking concentration

and the controls described above. We find, consistent with Result 3, a significant negative coeffi-

cient (t=-2.35) on banking concentration. A one-standard-deviation increase in bank concentration

reduces the level at which decisions are made by 2.2% of the mean.

These findings from the NOS data utilize only one proxy for financial constraints and draw on

a relatively small data set. The results, nonetheless, provide some empirical support for Result 3

and indicate that it has several clearly testable implications.

6 Conclusion

In this paper we develop a model of the optimal capital and labor management strategies of fi-

nancially constrained and unconstrained firms. We argue that workers are better informed than

suppliers of physical capital, and therefore employees, but not capital providers, have the real option

to abandon unsuccessful firms. As a result, employees are willing to work for firms without advance

payments, while capital suppliers are not, and constrained firms (that cannot offer initial payments)

consequently will have lower capital-labor ratios. We also demonstrate that successful constrained

firms find it difficult to hire new employees and thus offer an implicit retention guarantee to their

workers. As a result, employees in these firms will be likelier to make firm-specific investments.

We further show that constrained firms will exhibit declining relative labor productivity over time.

Empirical tests confirm the central predictions of the model and indicate that a firm’s use of labor

as an input in production can be affected considerably by the extent of its financial resources.

An interesting implication of the model is that firms that undergo a shift in their financial

status will simultaneously alter their labor policies. For example, the advent of new sizable venture

capital or bank financing may discourage employees from making firm-specific investments. In such

a setting, encouraging employee initiative becomes less important, which should lead the firm to

26

institute a more formal hierarchy structure that concentrates decision-making power in the hands

of senior managers. This may lead to a “professionalization” (Hellman and Puri (2002)) of the way

the firm conducts business that derives simply from the firm’s new access to financing and that is

independent of any actions or guidance on the part of the source of capital. Similarly, post-IPO

firms should experience a significant change in their employee retention policies.

References

Aghion, Philippe, and Jean Tirole, “Formal and Real Authority in Organizations,” The Journal ofPolitical Economy, 105 (1997), 1-29.

Almedia, Heitor, and Murillo Campello, “Financial Constraints, Asset Tangibility, and CorporateInvestment,” Working paper, New York University, (2005).

Angrist, J., “Estimation of Limited-Dependent Variable Models With Dummy Endogenous Regres-sors: Simple Strategies for Empirical Practice,” Technical Working Paper 248, National Bureau ofEconomic Research, (2000).

Audretsch, D., “New-Firm Survival and the Technological Regime”, Review of Economics andStatistics 73 (1991), 441-450.

Berger, Allen, Rosen, Richard, and Gregory Udell, “The Effect of Market Size Structure on Com-petition: The Case of Small Business Lending,” Federal Reserve Board Working Paper, (2001).

Binmore, Ken, Rubinstein, Ariel and Asher Wolinsky, “The Nash Bargaining Solution in EconomicModelling,” The RAND Journal of Economics 17 (1986), 176-188.

Black, Bernard S., and Ronald J. Gilson, “Venture Capital and the Structure of Capital Markets:Banks versus Stock Markets,” Journal of Financial Economics 47 (1998), 243-77.

Black, Sandra E., and Lisa M. Lynch, “How to Compete: The Impact of Workplace Practicesand Information Technology on Productivity,” The Review of Economics and Statistics 83 (2001),434445.

Black, Sandra E., and Philip E. Strahan, “Entrepreneurship and Bank Credit Availability,” Journalof Finance 57 (2002), 2807-33.

Bronars, Stephen G., and Donald R. Deere, “The Threat of Unionization, the Use of Debt, and thePreservation of Shareholder Wealth,” Quarterly Journal of Economics 106 (1991), 231-54.

Burgess, Robin, and Rohini Pande, “Do Rural Banks Matter? Evidence from the Indian SocialBanking Experiment,” Working paper, London School of Economics, (2003).

Carlin, Wendy, and Colin Mayer, “Finance, investment, and growth,” Journal of Financial Eco-nomics 69, 191-226.

Cetorelli, Nicola, and Michele Gambera, “Banking Market Structure, Financial Dependence andGrowth: International Evidence from Industry Data,” Journal of Finance 56 (2001), 617-648.

Cetorelli, Nicola, and Philip E. Strahan, “Finance as a Barrier to Entry: Bank Competition and

27

Industry Structure in Local U.S. Markets,” Working Paper, Federal Reserve Bank of Chicago,(2004).

DeMarzo, P. M., and Fishman, M. J., “Optimal Long-Term Financial Contracting with PrivatelyObserved Cash Flows, Working paper, Stanford University, (2002).

Demirguc-Kunt, Asli, and Vojislav Maksimovic, 1998, Law, Finance, and Firm Growth, Journal ofFinance 53, 2107-2137.

Dessein, Wouter, and Tano Santos, “The Demand for Coordination,” Working paper, ColumbiaUniversity Graduate School of Business, (2003).

Dupuy, A., and A. de Grip, “Do Large Firms Have More Opportunities to Substitute Labour ThanSmall Firms?,” Working paper, Maastricht University, (2003).

Eisfeldt, Andrea L., and Adriano A. Rampini, “Leasing, ability to repossess, and debt capacity,Working Paper, Northwestern University, (2005a).

Eisfeldt, Andrea L., and Adriano A. Rampini, “New or used? Investment with credit constraints,”Working Paper, Northwestern University (2005b).

Freeman, Richard, and Morris Kleiner, Who Benefits Most from Employee Involvement: Firms orWorkers? American Economic Review 90 (2000), 219-223.

Garicano, Luis, “Hierarchies and the Organization of Knowledge in Production,” Journal of Polit-ical Economy 108 (2000), 874-904.

Gromb, D., Renegotiation in Debt Contracts, Working Paper, MIT, (1999).

Hamilton, Barton H., “Does Entrepreneurship Pay? An Empirical Analysis of the Returns toSelf-Employment,” Journal of Political Economy 108(2000), 604-31.

Hanka, Gordon, “Debt and the Terms of Employment,” Journal of Financial Economics 48 (1998),245-82.

Hannan, Timothy H., “Market Share Inequality, the Number of Competitors, and the HHI: AnExamination of Bank Pricing,” Review of Industrial Organization, 12 (1997), 23-35.

Harris, Milt, and Artur Raviv, “The Role of Games in Security Design, Review of Financial Studies,8 (1995), 327-67

Harris, Milt, and Artur Raviv, “Organization Design,” Management Science 48 (2002), 852 - 865.

Hart, Oliver, and John Moore, “A Theory of Debt Based on the Inalienability of Human capital,”Quarterly Journal of Economics 109 (1994), 841-79.

Headd, Brian, “The characteristics of small-business employees,” Monthly Labor Review 123 (2000),13-18.

Hellmann, Thomas, and Manju Puri.“Venture Capital and the Professionalization of Start-UpFirms: Empirical Evidence,” Journal of Finance 57 (2002), 169-97

Inderst, Roman, and Holger M. Muller, “The Effect of Capital Market Characteristics on the Valueof Start-Up Firms,” Journal of Financial Economics 72 (2004), 319-56.

28

Joel Popkin and Company, “Small Business Share of Employment Growth,” report submitted tothe U.S. Small Business Administration, (2001).

Kalleberg, Arne L., David Knoke and Peter V. Marsden. National Organizations Survey (NOS),1996-1997 [Computer file]. ICPSR version. Minneapolis, MN: University of Minnesota Center forSurvey Research [producer], 1997. Ann Arbor, MI: Inter-university Consortium for Political andSocial Research [distributor], 2001.

Kahn, James, and Jong-Soo Lim, “Skilled Labor-Augmenting Technical Progress in U.S. Manufac-turing,” Quarterly Journal of Economics 113 (1998), 1281-1308.

Kang, Jun Koo, and Anil Shivdasani, “Corporate Restructuring during Performance Declines inJapan,” Journal of Financial Economics 46 (1997), 29-65.

Kelejian, Harry H., “Two-Stage Least Squares and Econometric Systems Linear in Parameters butNonlinear in the Endogenous Variables,” Journal of the American Statistical Association 66 (1971),373-374.

Kimura, Fukunari, “Subcontracting and the Performance of Small and Medium Firms in Japan,”Working paper, World Bank Institute, (2001).

Klein, Michael, Joe Peek and Eric Rosengren, “Troubled Banks, Impaired Foreign Investment: TheRole of Relative Access to Credit,” American Economic Review, 92 (2002), 664-682.

Lane, Julia, Allan Isaac and David Stevens, “Firm Heterogeneity and Worker Turnover,” Workingpaper, American University (1996).

Michelacci, Claudio, and Vincenzo Quadrini, “Financial Markets and Wages,” Working paper,University of Southern California (2005).

Moskowitz, Tobias J., and Annette Vissing-Jorgensen, “The Returns to Entrepreneurial Investment:A Private Equity Premium Puzzle?,” American Economic Review 92 (2002), 745-778.

Ofek, Eli, “Capital Structure and Firm Response to Poor Performance: An Empirical Analysis,”Journal of Financial Economics 34 (1993), 3-30.

Oi, Walter, Heterogeneous Firms and the Organization of Production, Economic Inquiry 21 (1983),14771.

Peek, Joe, and Eric Rosengren, “Collateral Damage: Effects of the Japanese Bank Crisis on RealActivity in the United States,” American Economic Review 90 (2000), 30-45.

Petersen, Mitchell, and Raghuram Rajan, “Does Distance Still Matter? The Information Revolu-tion in Small Business Lending,” Journal of Finance 57 (2002), 2533-2570.

Ueda, Masako, “Banks versus Venture Capital: Project Evaluation, Screening, and Expropriation,”Journal of Finance 59 (2004), 601-21.

Vayanos, Dimitri, “The Decentralization of Information Processing in the Presence of Interactions,”Review of Economic Studies 70 (2003), 667-95.

Wilkinson, Adrian, “Employment relations in SMEs,” Employee Relations 21 (1999), 206-217.

Wooldridge, Jefferey M., “Econometric Analysis of Cross Section and Panel Data,” MIT Press,Cambridge, MA (2002).

29

Appendix

Proof of Result 1.For each variable x, we denote its minimum and maximum by xmin and xmax, respectively. Suppose,for a contradiction, that there is an equilibrium in which the constrained entrepreneur hires capitalin the first period. We begin by showing that the constrained entrepreneur will retain the supplierof capital after first period production with probability one. It cannot be that the supplier ofcapital quits with probability one, since then (3) would show that this supplier would not supplythe entrepreneur in period 1. So for any outcome of f and r, the supplier of capital chooses toremain with positive probability ps > 0. (Recall that the supplier of capital cannot condition hisstrategy on f and r, since he does not view them until after making a decision.)Suppose that for any given f and r with some positive probability the entrepreneur fires the supplierof capital. It must then be that with some positive probability there is a supplier of capital willingto rent to the entrepreneur in the second period or else the entrepreneur would do strictly better toretain the first period supplier of capital. This implies that if a separation is viewed by the market,there is a probability pc > 0 that new capital is supplied. For any f and r, the entrepreneur’s payofffrom retention is (1 − θ)(f + r) and his payoff from firing the capital and attempting to rehire is(1− θ)(pc(f + r)). In equilibrium, the entrepreneur will choose to fire if and only if

(1− θ)(f + r) < (1− θ)pc(f + r),

i.e. r < pcr − (1 − pc)f . We define D(f) = min{rmax, pcr − (1 − pc)f}, a decreasing functionof f . Separation can arise from either quitting or firing. We denote the densities and cumulativedistribution functions of f , r and q by hf , hr and hq, Hf ,Hr and Hq, respectively. We have

E[f |separation] =ps

∫ fmax

fmin

∫ D(f)rmin

hr(r)dr(f)hf (f)df + (1− ps)f

ps

∫ fmax

fmin

∫ D(f)rmin

hr(r)drhf (f)df + (1− ps). (15)

We note that

∫ fmax

fmin

∫ D(f)rmin

hr(r)dr(f)hf (f)df∫ fmax

fmin

∫ D(f)rmin

hr(r)drhf (f)df=

∫ fmax

fminHr(D(f))(f)hf (f)df

∫ fmax

fminHr(D(f))hf (f)df

≤ f ,

since hf (f) = Hr(D(f))hf (f)R fmaxfmin

Hr(D(f))hf (f)dfis a probability density function for f , and it is dominated by

he in the sense of the monotone likelihood ratio property and hence has a lower mean. (It is clear

that hf (f)hf (f) = Hr(D(f))R fmax

fminHr(D(f))hf (f)df

is decreasing, since D is decreasing.) We conclude from (15) that

E[f |separation] ≤ f .

No new capital will consent to join the firm since the expected payoff from doing so is

θ(r + E[f |separation]) ≤ θ(r + f) < ζ.

This contradicts our earlier assumption. We conclude that with probability one the entrepreneurretains the first period supplier of capital. The two-period payoff to the first period supplier ofcapital is given by

θ(f + r)(1 + ps) + (1− ps)ζ < 2ζ.

30

We conclude that no supplier of capital will be willing to provide capital to the constrained entre-preneur in the first period.We now show that condition (4) is necessary for the existence of an equilibrium in which financiallyconstrained entrepreneurs hire labor. If in equilibrium labor is hired by the constrained entrepreneurin the first period, the worker will quit if

θ(f + q) < ε.

We denote the probability that new labor can be hired without any advance cost by pl. Theentrepreneur will fire the worker if

(f + q) < pl(f + q).

Separation thus occurs for

q < max{ ε

θ− f, pl(f + q)− f} := D1(f),

where the equality is a definition and D1 is a decreasing function. We thus have

E[f |separation] =

∫ fmax

fmin

∫ D1(f)qmin

hq(q)dq(f)hf (f)df∫ fmax

fmin

∫ D1(f)qmin

hq(q)dqhf (f)df. (16)

As above, this shows that after separation no new labor will be supplied to the constrained entre-preneur. Since no new labor can be hired in period two, the first period worker will remain withthe constrained firm if and only if θ(q + f) ≥ ε (i.e. if it is in his own interest to do so). The twoperiod expected payoff to a worker hired by a constrained entrepreneur in the first period is

θ(f + q) + θE[max{q + f,ε

θ}] ≥ 2θ(f + q).

If

θ(f + q) + θE[max{q + f,ε

θ}] ≥ 2ε, (17)

it is an equilibrium for a worker to join the firm in period 1, quit if θ(q+f) < ε and remain otherwise.The entrepreneur never fires the worker. All off-equilibrium path actions from the entrepreneur areassumed to come from entrepreneurs with f = fmin in the second period. Condition (17) is thusnecessary and sufficient for the existence of an equilibrium in which labor is hired, since if it failsthe worker will not agree to join the firm and if it holds he will.Proof of Result 2.To prove the result we will exhibit a set of equilibria in which the unconstrained entrepreneurproduces using capital and a second set of equilibria in which he produces using labor. All referencesto an entrepreneur below are to an unconstrained entrepreneur.First, we describe an equilibrium in which the entrepreneur produces using capital. We assume

r > Max{2q,θfmin

1− θ+ q, ζ + 2(1− θ)q} (18)

and

ζ − θ(f + r) ∈ (0, min{ε− θ(f + q), θf}). (19)

31

These assumptions guarantee that capital is productive relative to labor (18) and not too expensive(19). We propose the following candidate equilibrium. Entrepreneurs hire capital in the first periodand keep it if and only if r ≥ r. Otherwise, the existing capital is sold and new capital is purchased.All off-equilibrium path bids from the entrepreneur are assumed to come from entrepreneurs withf = fmin and employees (who are never hired in equilibrium) expect not to be retained.We now confirm that the above describes an equilibrium. Assumptions (18) and (19) show thatr − ζ ≥ q − ε, so hiring labor in the first round and then buying capital is dominated by buyingcapital in both rounds. Given the off-equilibrium path beliefs about entrepreneurs who hire labor,a similar argument shows that if capital was hired in the first period, buying new capital is superiorto hiring new labor in the second period. The off-equilibrium path beliefs also make buying capitalalways cheaper than renting it. Condition (18) shows that the entrepreneur does better to followhis equilibrium strategy rather than not producing.We last consider the strategy of using labor in both periods. The expected first-period payoff fromthis strategy is (1 − θ)(f + q). If labor is retained and θ(f + q) > ε, then no retention paymentis needed. If (f + q) ∈ [ε, ε

θ ), a retention payment of kl = ε − θ(f + q) will be required. Theentrepreneur will never retain the labor if (f + q) < ε, since he could only do so at a loss. Thisgives a net payoff to the entrepreneur choosing retention of

Max{Min{f + q − ε, (1− θ)(f + q)}, 0}.If the entrepreneur hires new labor, the off-equilibrium beliefs imply that only signing paymentss ≥ ε− θq− θfmin will be accepted. The expected payoff from hiring new labor is (1− θ)(f + q)−(ε− θq − θfmin). The two-period expected payoff to the entrepreneur hiring labor is thus

(1− θ)(f + q) + E [Max{(1− θ)f + q − ε + θfmin,Min{f + q − ε, (1− θ)(f + q)}}]

≤ (1− θ)(f + q) + E [Max{(1− θ)f + r − ζ, (1− θ)(f + q)}]

≤ 2(1− θ)(f + q) + r − ζ

≤ 2(f + r − ζ)

≤ 2(f + r − ζ) + E [Max{r − r, 0}] ,where the last term is the expected payoff from following the candidate equilibrium strategy ofusing capital in both periods. The first inequality follows from (18) and (19), the third from (18),and the second and fourth follow from routine algebra. This verifies the candidate equilibrium.We now describe an equilibrium in which the entrepreneur produces using labor. The necessaryconditions are

q > Max{ε, θfmin

1− θ+ r, 2r,

θf + r − ζ2

1− θ}, (20)

ε− θ(f + q) < ζ − θ(f + r), (21)

and

32

ε− θ(f + q) ∈ (0, E

[max{ε, θ(f + q)I(1−θ)q≥q−ε+θfmin

}]− ε). (22)

Condition (22) implies (4).We propose the following candidate equilibrium. Entrepreneurs hire labor at no cost in the firstperiod. The retention payment to first period labor is kl = Max{ε − θ(f + q), 0} and the signingpayment to new second period labor is s = ε − θq − θfmin. First period labor is retained by theentrepreneurs if and only if Min{f + q − ε, (1 − θ)(f + q)} ≥ max{(1 − θ)f + q − ε + θfmin, 0};otherwise new labor is hired (if the first argument of the maximum is larger) or the firm closes (if thesecond argument is larger). All off-equilibrium path bids are assumed to come from entrepreneurswith f = fmin in the second period. The proof that this is an equilibrium is analogous to thatgiven above for the physical capital equilibrium.

Proof of Lemma 1.The entrepreneur at the end of the first period has potentially three actions from which to choose:1. σa - retain the employee:

πE(f, q, σa) = min{(1− θ)(f + q), f + q − ε}.2. σb - fire the employee and hire a replacement:

πE(f, q, σb) = (1− θ)(f + q)− s,

where s is the signing payment.3. σc - fire the employee leave the position open:

πE(f, q, σc) = 0.

The proof of Result 1 showed that E[f |separation] ≤ f for the constrained entrepreneur, whichimplies that the labor market will demand a positive signing payment s > 0 for joining the firmat the end of the first period. The constrained entrepreneur cannot pay this amount, so σb isunavailable to him. If θ(f + q) > ε, then σa dominates σc.Proof of Result 3.We define R to be the indicator function for the retention of the worker: R(f, q) = 1 if the workeris retained by the firm when firm productivity is given by f and end-of-first-period worker qualityis given by q, and R(f, q) = 0 otherwise. Rc and Ru are the indicator functions for the constrainedand unconstrained firms, respectively.The worker may always choose to quit and receive ε, so for a given firm-wide quality f and end-of-first-period worker quality q, the expected second period payoff of a worker in a constrained firmis given by

W 2,c(f, q) = Rc(f, q)max{θ(f + q), ε}+ (1−Rc(f, q)) ε. (23)

The formula for W 2,u(f, q) is analogous, with unconstrained retention functions replacing the con-strained ones above.The first statement in Result 3 is that if

W 2,u(f, q + λ)−W 2,u(f, q) ≥ g − θλ, (24)

then

33

W 2,c(f, q + λ)−W 2,c(f, q) ≥ g − θλ. (25)

First suppose that θ(e + q + λ) ≤ ε. This implies that W 2,u(e, q + λ) = ε = W 2,u(e, q), henceinequality (5) shows that (24) fails and the result holds.Now suppose that θ(e + q + λ) > ε. Lemma 1 shows that R2,c(e, q + λ) = 1. We therefore haveW 2,c(e, q + λ) = θ(e + q + λ) and W 2,u(e, q + λ) ≤ θ(e + q + λ). If θ(e + q) ≤ ε then W 2,c(e, q) = ε

and W 2,u(e, q) = ε, (24) implies (25) and the result is proved.Suppose θ(e + q) > ε. This implies that

W 2,c(e, q + λ)−W 2,c(e, q) ≥ θ(e + q + λ)− θ(e + q)

≥ g − θλ,

where the final inequality follows from (6). This shows that (25) holds.For the failure of the converse, consider the labor-hiring equilibrium described in the proof of Result2. If q ∈ (q−2λ− s

1−θ , q−λ− s1−θ ) and θ(f +q)+2θλ−g > ε then the worker makes a firm-specific

investment in the constrained firm, but not in the unconstrained firm.

Proof of Result 4.For any given f and first-period worker quality q, Result 3 shows that if firm-specific investmentoccurs in the unconstrained firm then it also occurs in the constrained firm. That shows thatE

[π1,c

] ≥ E[π1,u

]. We denote the expected second period worker match quality by qc and qu, for

the constrained and unconstrained entrepreneurs, respectively. If first-period firm-specific invest-ment occurs in either both the constrained and unconstrained firms or in neither, at the end ofthe first period the employee in both firm types has the same quality q. As described in the proofof Lemma 1, the unconstrained entrepreneur has more second-period options from which to pick.The additional options involve making retention payments (σa) or signing payments (σb), so theunconstrained entrepreneur will only select them if qu ≥ qc (and, hence, π2,u ≥ π2,c). Otherwise,the two entrepreneurs choose the same action and qu = qc and π2,u = π2,c.

The second case is the one in which first-period investment occurs only in the constrained firm.In this case (25) holds, and the proof of Result 3 shows that it must be that θ(f + q + λ) > ε.This shows that action σc is dominated for the unconstrained entrepreneur. We first suppose thatπE(f, q + λ, σb) < πE(f, q + λ, σa). That implies that the unconstrained entrepreneur retains thefirst-period worker, so W 2,u(f, q + λ) = W 2,c(f, q + λ). We also have W 2,u(f, q) ≤ W 2,c(f, q) ingeneral because the additional options available to unconstrained entrepreneurs grant the employeehis minimum payoff of ε. This shows that (25) implies (24), so the employee in the unconstrainedfirm would also make a firm-specific investment, which contradicts our initial assumption. We nowsuppose that πE(f, q+λ, σb) ≥ πE(f, q+λ, σa). That implies that (1−θ)(f+q)−s ≥ (1−θ)(f+q+λ).We conclude that qu = q ≥ q + λ = qc, so π2,u ≥ π2,c in this second case as well.

We now prove the second part of the result about second-period conditional output. If x < εθ , then

the employee in the constrained firm will leave after the first period, since his payoff in the firm willnot increase in the second period and he does better to receive his opportunity cost. In this case,since the constrained firm cannot hire a new employee, E

[π2,c|π1,c = x

]= 0 ≤ E

[π2,u|π1,u = x

].

If x ≥ εθ , then E

[π2,c|π1,c = x

] ≤ x, since the constrained firm can do no better than to retainthe existing employee, and this employee will never make a second-period firm-specific investment.Given a first-period output of x ≥ ε

θ , the unconstrained entrepreneur may either retain the employee

34

or hire a new one, so his expected second-period output is weakly higher than x. This shows thatE

[π2,c|π1,c = x

] ≤ x ≤ E[π2,u|π1,u = x

].

35

Table 1: Summary Statistics

National Survey of Small Business Finance variable distribution characteristicsStd

Non-binary variables Mean Median Dev Binary variables MeanEmployees 25.5 5 54.52 MSA 0.78Assets 1,463,454 95,756 5,330,001 Female 0.22Depreciable Assets 419,825 17,577 2,114,480 African-American 0.08Sales 3,436,004 250,000 14,993,484 Hispanic 0.07Total Costs 3,046,484 195,000 13,882,678 Applied for a loan 0.27Profit Margin -0.03 0.13 4.89 Applied for a loan and rejected 0.05Bank Herf Index 2.47 3 0.59Home Equity 147,677 80,000 323,491Net Worth- Home Equity 786,430 180,000 3,425,042DBscore 3.03 3 1.04Education 4.65 6 1.96Firm Age 14.45 11 12.12

Table 1 reports means (all variables), medians and standard deviations (all non-binary variables) for each variable

in the 1998 National Survey of Small Business Finance. Employees, assets, depreciable assets (including depletable

and intangible assets), sales, total costs of conducting business, firm age, profit margin (profits divided by sales) and

DBscore (Dun and Bradstreet credit score on a 1-5 scale) are measured at the firm level. Home equity, net worth

(excluding the value of the firm and the value of home equity), education (on a 1-7 scale), gender and ethnicity are

given for the firm’s primary owner. MSA is a dummy for whether the firm is located in a Metropolitan Statistical

Area, and Bank Herf Index is a measure (from 1 to 3) of the deposit concentration of the local MSA or county

commercial bank market. Applied for a loan is a dummy for whether the firm made one or more loan applications in

the past 3 years, and Applied for a loan and rejected is a dummy for whether the firm’s most recent loan application

(if any) was rejected.

36

Table 2: The Capital-Labor Decision and Financial Constraints

(1) (2) (3) (4) (5) (6)Dependent Loan appl. Capital-Labor Capital-Labor Capital-Labor Capital-Labor Capital-Labor

Variable rejected? ratio ratio ratio ratio ratio# Obs. 952 952 952 436 516 952

Bank Herf Index 0.497** -0.143*(2.36) (-1.69)

Log(home equity) -0.086** 0.044**(-3.08) (2.70)

Log(net worth - home eq.) -0.064** 0.037**(-2.07) (2.01)

Education -0.085 0.033(-1.28) (1.16)

Female 0.219 -0.262*(0.76) (-1.72)

African-American 1.871** -0.563**(5.58) (-2.15)

Hispanic 1.037** 0.034(2.98) (0.17)

DBscore -0.202* 0.073*(-1.94) (1.79)

Log(1+sales) -0.377** 0.207** 0.160** 0.237** 0.239** 0.274**(-4.87) (4.13) (2.26) (3.64) (3.34) (6.91)

Log(1+firm age) -0.097 0.104 0.083 -0.109 0.246 0.133(-0.52) (1.00) (0.73) (-0.47) (1.25) (1.36)

Profit margin -0.084** -0.100** -0.113** -0.104** 0.071 -0.051*(-2.16) (-3.08) (-3.01) (-3.19) (0.50) (-1.89)

MSA? 0.652** 0.049 0.104 0.051 0.019 -0.145(2.22) (0.38) (0.72) (0.31) (0.10) (-1.20)

Fin. Constr. (Inst.) -2.344** -1.332 -2.159**(-3.93) (-1.22) (-3.58)

Fin. Constr. (Inst.’) -3.200**(-3.15)

Estimation Method Logit 2SLS 2SLS 2SLS 2SLSSubsample? NA NA NA Old firms Young firms

R2 0.31 0.16 0.01 0.33 0.24

Results from the regressions of whether the firm had a loan application declined in the last 3 years (first column) and

the log of one plus the ratio of assets to employees (second through sixth columns) on measures of financial constraints.

The regressors with reported coefficients are an index of the local banking market Herfindahl measure, the log of one

plus the firm owner’s home equity, the log of one plus the owner’s net worth excluding his home equity, a ranking of the

firm owner’s education, dummies for the firm owner being female, African-American or Hispanic, the firm’s credit score

as obtained from Dun and Bradstreet, the log of one plus annual sales, the log of one plus firm age, the firm’s profits to

sales ratio (profit margin), a dummy for the location of the firm within an MSA (Metropolitan Statistical Area) and an

instrumented estimate of the probability of the firm’s having a loan application declined. The instruments are the first

eight variables listed for the regressions in columns two, four and five, and the first three variables listed for the regression

in column three. Additional regressors with unreported coefficients include a constant, industry dummies at the 2-digit

SIC level, census region dummies, the log of one plus the owner’s experience in managing a business, the log of the owner’s

age, the ownership share of the primary owner, a dummy for owner-managed firms, a dummy for firms with zero sales in

the previous year, a dummy for new firms (except column four), a dummy for firms subject to corporate income tax and a

dummy for family-owned firms. The data source is the 1998 National Survey of Small Business Finance. The regressions

are estimated via binary logistic regression (Logit), ordinary least squares (OLS), or two-stage least squares (2SLS), as

described, with t-statistics reported in parentheses using robust standard errors. Reported R2 for the Logit specification

is McFadden’s pseudo R2.

*,** Indicates significance at the 10% and 5% levels, respectively.

37

Table 3: The Labor Intensity of Production and Financial Constraints- Robustness Tests

(1) (2) (3) (4)Dependent Total Costs-Labor Total Costs-Labor Depreciable Assets-Labor Depreciable Assets-Labor

Variable ratio ratio ratio ratio# Obs. 948 948 951 951

Fin. Constr. (Inst.) -1.042** -2.843**(-3.68) (-2.60)

Fin. Constr. (Inst.’) -1.053** -4.547**(-2.16) (-2.37)

Log(1+sales) 0.325** 0.324** 0.333** 0.238*(11.79) (8.76) (3.57) (1.81)

Log(1+firm age) -0.093 -0.093 -0.002 -0.043(-1.52) (-1.50) (-0.01) (-0.21)

Profit margin -0.200** -0.226**(-3.94) (-3.84)

MSA? 0.150* 0.151* -0.617** -0.506(1.85) (1.72) (-2.13) (-1.59)

Estimation Method 2SLS 2SLS 2SLS 2SLSR2 0.48 0.47 0.13 0.03

Results from the regressions of the log of one plus the ratio of the total cost of conducting business to the number of

employees (first and second columns) and the log of one plus the ratio of depreciable, depletable or intangible assets to

employees (third and fourth columns) on measures of financial constraints. The regressors with reported coefficients

are an instrumented estimate of the probability of the firm’s having a loan application declined, the log of one plus

annual sales, the log of one plus firm age, the firm’s profits to sales ratio (profit margin) in columns one and three and

a dummy for the location of the firm within an MSA (Metropolitan Statistical Area). The instruments are generated

using an index of the local banking market Herfindahl measure, the log of one plus the firm owner’s home equity, the

log of one plus the owner’s net worth excluding his home equity, a ranking of the firm owner’s education, dummies

for the firm owner being female, African-American or Hispanic and the firm’s credit score as obtained from Dun and

Bradstreet for the regressions in columns one and three, and are generated using the first three of these variables

for the regressions in columns two and four. Additional regressors with coefficients unreported for brevity include a

constant, industry dummies at the 2-digit SIC level, census region dummies, the log of one plus the owner’s experience

in managing a business, the log of the owner’s age, the ownership share of the primary owner, a dummy for whether

the firm is managed by the owner, a dummy for firms with zero sales in the previous year, a dummy for new firms,

a dummy for firms subject to corporate income tax and a dummy for family-owned firms. The data source is the

1998 National Survey of Small Business Finance. The regressions are estimated via ordinary least squares (OLS) or

two-stage least squares (2SLS), as described, with t-statistics reported in parentheses using robust standard errors.

*,** Indicates significance at the 10% and 5% levels, respectively.

38

Table 4: Productivity and Financial Constraints

(1) (2) (3) (4)Dependent

Variable Log(sales) Log(sales) Log(sales) Log(sales)# Obs. 436 516 436 516

Fin. Constr. (Inst.) 0.514 0.524(0.42) (0.89)

Fin. Constr. * Log(1+firm age) (Inst.) -0.389 -0.629**(-0.98) (-2.12)

Log(1+firm age) 0.059 0.345**(0.45) (3.61)

Fin. Constr. (Inst.’) 0.113 0.722(0.07) (1.23)

Fin. Constr. * Log(1+firm age) (Inst.’) -0.182 -0.712**(-0.34) (-2.31)

Log(1+firm age) 0.049 0.365**(0.38) (3.87)

MSA? 0.222** 0.060 0.207** 0.059(2.70) (0.67) (2.51) (0.65)

Log(1+assets) * Industry dummies Yes Yes Yes YesLog(emp) * Industry dummies Yes Yes Yes YesAsset category controls Yes Yes Yes YesEstimation Method 2SLS 2SLS 2SLS 2SLSSubsample? Old firms Young firms Old firms Young firms

R2 0.92 0.88 0.92 0.88

Results from the regressions of the log of annual sales on measures of financial constraints, interactions between

financial constraints and firm age and firm inputs. The regressors with reported coefficients are an instrumented

estimate of the probability of the firm’s having a loan application declined, the interaction of the instrumented

estimate with the log of one plus the firm age, the log of one plus the firm age and a dummy for the location of

the firm within an MSA (Metropolitan Statistical Area). The instruments are generated using an index of the local

banking market Herfindahl measure, the log of one plus the firm owner’s home equity, the log of one plus the owner’s

net worth excluding his home equity, a ranking of the firm owner’s education, dummies for the firm owner being female,

African-American or Hispanic and the firm’s credit score as obtained from Dun and Bradstreet for the regressions in

columns one and two, and are generated using the first three of these variables for the regressions in columns three

and four. Additional regressors with coefficients unreported for brevity include a constant, industry dummies at the

2-digit SIC level, industry dummies interacted with both the log of one plus assets and the log of employees, the log

of one plus the asset type for the following eight asset categories: cash, accounts receivable, inventory, other current

assets, other investments, land, buildings plus equipment and other assets, census region dummies, the log of one

plus the owner’s experience in managing a business, the log of the owner’s age, the ownership share of the primary

owner, a dummy for whether the firm is managed by the owner, a dummy for firms subject to corporate income tax,

a dummy for family-owned firms and an index for the geographic size of the firm’s market. The data source is the

1998 National Survey of Small Business Finance. The regressions are estimated via two-stage least squares (2SLS),

with t-statistics reported in parentheses using robust standard errors.

*,** Indicates significance at the 10% and 5% levels, respectively.

39

The entrepreneur hires either labor or capital.

The entrepreneur and labor (if hired) view firm quality fand either labor match quality q or capital match quality r.

The entrepreneur may fire the supplier of labor or capital. The input supplier may also terminate the relationship.

If both parties choose to maintain the relationship, it continues into the second period. Otherwise, the entrepreneur may attempt to hire new capital or labor.

First period production ends and the output is divided.

Second period timing is analogous, with the exception that no arrangements are made for future production.

First period production begins.

t=0 t=1 t=2

Second period production begins.

40

The entrepreneur hires labor.

The entrepreneur and labor view firm quality f and labor match quality q.

The entrepreneur may fire the worker and the worker may quit.

If both parties choose to maintain the relationship, it continues into the second period. Otherwise, the entrepreneur may attempt to hire new labor.

First period production ends and the output is divided.

Second period timing is analogous, with the exception that no arrangements are made for future production.

First period production begins.

t=0 t=1 t=2

Second period production begins.

The worker decides whether or not to make a firm-specific investment.

41